The connectome-based predictive modeling (CPM) was used to examine the association between bingeing with connectivity edges. The protocol was based on a published methodology (Shen et al., 2017) and was run separately for standard and extreme bingeing. Input data were composed of change in connectivity edges (time 2 minus time 1), longitudinal standard/extreme bingeing, and covariate variables including sex, SES, group, scanner (same or different scanners across the two sessions), and change in age and FD (time 2 minus time 1). These input data were split into training/test sets using leave-one-out cross validation. The CPM protocol involved the following steps: 1) select important edges that demonstrate significant associations with bingeing using Spearman partial correlation, 2) fit a model using the training set and apply this model to predict bingeing from change in connectivity edges in the test set, 3) calculate model accuracy using the test sets across all cross-validation folds, and 4) determine if this model accuracy is significant using permutation test. Further details of the CPM protocol are provided in the Supplementary Materials.
